基於蒙地卡羅樹搜尋與多目標優化的自適應樓層規劃

Abstract

本研究提出一項稱為AdaptMCTS（Adaptive Monte Carlo Tree Search）的自適應樓層規劃演算法，針對多樓層建築中的房間配置問題，結合蒙地卡羅樹搜尋法與多目標優化技術，於有限時間內快速產生多組高品質可行解。AdaptMCTS 透過階層式樹結構與模擬機制處理樓層容量限制、地形限制及房型鄰近性等實務條件，並在模擬時具備自動回溯與重試功能，有效提升求解穩定性與覆蓋率。此外，本方法整合房型緊密度、類型分佈與空間利用率等多目標指標，建立量化評估函數以篩選優解，並具備並行搜索能力，可於有限時間內產出多樣化的可行配置方案，未來亦可作為機器學習模型訓練資料之來源，有助於解決實務上歷史資料取得不易的問題。實驗以多種樓層數與不同地形情境進行測試，結果顯示AdaptMCTS 在解的品質與擴展性方面均優於模擬退火法與整數規劃法，特別是在高維度或條件頻繁變動的場景下具備更佳表現。綜合而言，AdaptMCTS 為解決複雜建築空間配置問題提供一種具彈性、穩定性與實務應用潛力的優化方法。

This thesis presents AdaptMCTS, an adaptive floor planning algorithm based on Monte Carlo Tree Search (MCTS) for optimizing multi-story room layouts. The proposed approach integrates MCTS with multi-objective heuristics to address practical constraints, including floor capacity, spatial adjacency, and terrain feasibility. A hierarchical search structure and parallel sampling strategy are employed to improve exploration efficiency and enhance solution diversity within limited runtime. A composite scoring function evaluates candidate layouts based on spatial compactness, type clustering, and space utilization. Experimental results show that AdaptMCTS outperforms Simulated Annealing (SA) and Integer Linear Programming (ILP) in both solution quality and scalability, especially under high-dimensional and dynamic conditions.